A risk-informed urban low-altitude air network planning method under large language model guidance

The rapid expansion of urban air mobility calls for safe low-altitude corridor networks across complex urban airspace, where heterogeneous positioning uncertainties caused by navigation and communication errors elevate collision risk and constrain both route feasibility and network-wide connectivity. To address this issue, this study proposes a risk-informed network planning framework that integrates three-dimensional collision risk modeling with language-model-guided multi-agent learning for the construction of safe urban air corridor networks. Specifically, a three-dimensional collision risk map is first constructed by translating navigation and communication errors into spatially varying positioning uncertainty, providing a unified safety basis for route feasibility evaluation. Network construction is then formulated as a cooperative graph connectivity task on the risk map, in which multiple learning agents jointly balance route connectivity, construction cost, and accumulated collision risk through coordinated training and distributed execution. To further accelerate convergence in complex urban environments, a vision-language model interprets airspace structure and produces high-level priors that steer agents toward safer altitude regimes. A case study over the low-altitude airspace of Nanjing demonstrates the effectiveness of the proposed framework in producing risk-compliant air corridor networks, with language-model-derived priors further reducing training effort by 50.3 %. The results support safer and more systematic air network strategies in complex urban environments, contributing to evidence-based collision risk mitigation and airspace management in the emerging urban low-altitude economy.